Ratings & Reviews Analytics: Star Distributions, Fair Average Ratings, Confidence-Adjusted Rankings, Review Velocity & Response Rates

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Count the reviews per star for one product, turn the counts into shares, and compare the average rating with the median – and see why a J-shaped distribution makes the average look worse than the typical review.
• Calculate the catalogue's weighted average rating, give every product a fair (Bayesian) average that pulls small samples towards it, and compare the ranking by raw average with the ranking by fair average.
• Calculate the share of positive reviews per product, its Wilson lower bound – the share you can be fairly sure of – and rank the products by that bound, so a perfect 8 out of 8 no longer beats 801 out of 900.
• Count reviews per product and week, measure how many days have passed since each product's last review, flag products whose reviews have gone stale and fit a trend to six months of review counts.
• Flag size problems in single reviews with FIND, then compare the average stars of reviews that mention delivery with those that do not, and measure which themes appear in the 1–2-star reviews.
• Calculate the response time of every public reply in hours, check it against a 24-hour target, measure how many critical reviews got an answer and find the median response time.
• Calculate each product page's conversion rate, sort products into rating bands, compare the pooled conversion of each band and fit a slope that shows how conversion moves with the rating.
• Build Hollowbrook's reputation dashboard: response rate and fair average per location, a traffic-light status from both, and a chart of the average stars by location.